Multiple Levels Comes Early?

[rdstern]

This is also to be sent to Lucie (and David) in case it affects their course. Here I particularly consider the R-Instat implications - that's if you agree?

I suggest for R-Instat the implications are modest and simple, because it is "built" for this.

Most data are presented as single data frames, e.g. diamonds data. I suggest we add versions of our data at multiple levels, where appropriate, in the R-Instat library. This to start with:

1. rice survey data (36 observations) where we add a version that has a data frame with variey summaries, one with village summaries and one at the variety by village level.
2. dodoma data where we add a data frame at the monthly level (for each year), the yearly level, and the month level (just 12 rows)
3. Similarly for the Ghana climatic data.

In the olden days I was concerned that many examples were provided as summary data, (e.g. contingency table) because that was the level for the proposed analysis (or significance test). I was keen to start with the primary data. Now I am keen to show both, as a routine.

I wonder about our teaching? For example where is this in Lucie's proposed course? An example is perhaps an on-farm trial that includes 10 varieties of millet. We have plot-level data with a column for the variety, and another for the farmer and a third for the yield. Could we also encourage a variety level table, and a farmer, level table. At the simplest, they just include the varieties and farmers in the trial, (and summaries from the plot-level data), but they might also include more farmers, from which those for the trial were selected, and possibly with more information about them. Similarly for the varieties.

Where is this in your current course, or is it in the second course? Where should it be?